NeurIPS 2025 Papers — Page 30
Conference on Neural Information Processing Systems · 5275 papers
MOOSE-Chem2: Exploring LLM Limits in Fine-Grained Scientific Hypothesis Discovery via Hierarchical Search
Zonglin Yang (Nanyang Technological University), Dongzhan Zhou (Shanghai Artificial Intelligence Laboratory)
OptimizationDrug DiscoveryTransformerLarge Language ModelAgentic AITextBenchmark
🎯 What it does: This paper proposes a fine-grained scientific hypothesis generation task, modeling it as a combinatorial optimization problem, and designs a Hierarchical Heuristic Search (HHS) method to implement it.
MoPFormer: Motion-Primitive Transformer for Wearable-Sensor Activity Recognition
Hao Zhang (Southern University of Science and Technology), Yu Zhang (Southern University of Science and Technology)
RecognitionExplainability and InterpretabilityTransformerContrastive LearningTime Series
🎯 What it does: Designed and evaluated MoPFormer—a Transformer framework based on motion primitives for action recognition using wearable IMU sensors.
More of the Same: Persistent Representational Harms Under Increased Representation
Jennifer Mickel (University of Texas at Austin), Kevin Tian (University of Texas at Austin)
GenerationRepresentation LearningTransformerLarge Language ModelPrompt EngineeringText
🎯 What it does: This paper proposes the GAS(P) evaluation method, which reveals the representation differences of different groups in generated texts without explicitly prompting group identities, and conducts empirical research in the context of gender and occupation.
More Than Generation: Unifying Generation and Depth Estimation via Text-to-Image Diffusion Models
Hongkai Lin (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
GenerationDepth EstimationTransformerDiffusion modelImage
🎯 What it does: In a fixed text-to-image diffusion model, a pluggable transformer is utilized to achieve a unification of generation and depth estimation tasks;
More Than Just Functional: LLM-as-a-Critique for Efficient Code Generation
Derui Zhu (Technical University of Munich), Weiyi Shang (University of Waterloo)
OptimizationAI Code AssistantTransformerLarge Language ModelReinforcement LearningPrompt EngineeringText
🎯 What it does: Utilizing prompts to enable large language models (LLMs) to serve as 'efficiency judges', guiding the generation of more efficient programs through a reward mechanism during code generation, without the need for code execution verification.
More Thinking, Less Seeing? Assessing Amplified Hallucination in Multimodal Reasoning Models
Zhongxing Xu (Stanford University), Sheng Liu (Stanford University)
Reinforcement LearningVision Language ModelMultimodalityBenchmark
🎯 What it does: This paper studies the balance between the length of reasoning chains and visual hallucinations in multimodal reasoning models, and systematically evaluates the amplifying effect of reasoning chains on hallucinations.
MoRE-Brain: Routed Mixture of Experts for Interpretable and Generalizable Cross-Subject fMRI Visual Decoding
YUXIANG WEI, Vince Calhoun
GenerationExplainability and InterpretabilityMixture of ExpertsDiffusion modelBiomedical DataMagnetic Resonance Imaging
🎯 What it does: This study proposes a mixed expert network based on brain functional hierarchy (MoRE-Brain), which maps fMRI signals to CLIP space and achieves high-fidelity, interpretable visual reconstruction through a dynamic time/space routing condition-controlled diffusion model.
MoRIC: A Modular Region-based Implicit Codec for Image Compression
Gen Li (Imperial College London), Deniz Gunduz
CompressionAuto EncoderImage
🎯 What it does: This paper proposes a region-based implicit encoder MoRIC, which achieves efficient compression by segmenting images into several regions and training dedicated local implicit neural representations (INR) for each region, supporting single-object compression and scalable layered compression.
MOSDT: Self-Distillation-Based Decision Transformer for Multi-Agent Offline Safe Reinforcement Learning
Yuchen Xia (Chinese University of Hong Kong), Yunjian Xu (Chinese University of Hong Kong)
Knowledge DistillationTransformerReinforcement LearningSequentialBenchmark
🎯 What it does: The first algorithm for Multi-Agent Offline Safe Reinforcement Learning (MOSRL) called MOSDT is proposed, along with the first MOSRL dataset and benchmark, MOSDB.
Motion Matters: Compact Gaussian Streaming for Free-Viewpoint Video Reconstruction
Jiacong Chen (Shenzhen University), Yongsheng Liang (Shenzhen University)
Data SynthesisCompressionComputational EfficiencyGaussian SplattingVideo
🎯 What it does: A Compact Gaussian Streaming (ComGS) framework is proposed, utilizing keypoint-driven motion representation to achieve efficient real-time reconstruction and transmission of online free-viewpoint video (FVV).
MOTION: Multi-Sculpt Evolutionary Coarsening for Federated Continual Graph Learning
Guancheng Wan (Wuhan University), Mang Ye (Wuhan University)
Federated LearningGraph Neural NetworkMixture of ExpertsGraph
🎯 What it does: This paper studies the problem of Federated Continual Graph Learning (FCGL) and proposes the MOTION framework, which retains historical graph topology using Multi-Expert Trimming (G-TMSC) on the client side and mitigates parameter conflicts using Topology-Sensitive Compatible Matrix Adaptive Aggregation (G-EPAE) on the server side.
Motion4D: Learning 3D-Consistent Motion and Semantics for 4D Scene Understanding
Haoran Zhou (National University of Singapore), Gim Hee Lee (National University of Singapore)
Object TrackingSegmentationGaussian SplattingVideo
🎯 What it does: Developed the Motion4D framework, which utilizes 2D priors to fuse 4D Gaussian Splatting for motion and semantic consistency in dynamic 3D scenes.
MotionBind: Multi-Modal Human Motion Alignment for Retrieval, Recognition, and Generation
Kaleab A Kinfu, Rene Vidal
RecognitionGenerationRetrievalTransformerDiffusion modelContrastive LearningVideoTextMultimodalityAudio
🎯 What it does: Incorporating human motion into a multimodal learning framework, a unified embedding space for motion and modalities such as text, images, and audio is constructed, enabling cross-modal retrieval, zero-shot action recognition, and generation from any modality to motion.
MotionRAG: Motion Retrieval-Augmented Image-to-Video Generation
Chenhui Zhu (Nanjing University), Limin Wang (Nanjing University)
GenerationRetrievalTransformerDiffusion modelImageVideoRetrieval-Augmented Generation
🎯 What it does: By retrieving motion patterns from relevant videos and transferring them to the input image, more realistic motion generation is achieved.
Mozart: Modularized and Efficient MoE Training on 3.5D Wafer-Scale Chiplet Architectures
Shuqing Luo (University of North Carolina at Chapel Hill), Tianlong Chen (University of North Carolina at Chapel Hill)
Large Language ModelMixture of ExpertsText
🎯 What it does: The Mozart framework is proposed, which implements an efficient post-training process for training large language models using Mixture-of-Experts (MoE) on a 3.5D wafer-scale chiplet architecture.
MPCache: MPC-Friendly KV Cache Eviction for Efficient Private LLM Inference
Wenxuan Zeng (Peking University), Meng Li (Peking University)
Safty and PrivacyComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: This paper proposes MPCACHE, a KV cache eviction framework designed for secure multi-party computation, aimed at enabling private LLM inference.
MPMAvatar: Learning 3D Gaussian Avatars with Accurate and Robust Physics-Based Dynamics
Changmin Lee (KAIST), Tae-Kyun Kim (KAIST)
GenerationVideoMeshPhysics Related
🎯 What it does: Through multi-view video learning, a system is constructed to perform physically accurate dynamic simulations of clothing using the Material Point Method (MPM) and combined with 3D Gaussian spring rendering, capable of generating realistic 3D human avatars from any perspective.
MPS-Prover: Advancing Stepwise Theorem Proving by Multi-Perspective Search and Data Curation
Zhenwen Liang (Tencent), Dong Yu (Tencent)
Large Language ModelTextBenchmark
🎯 What it does: A stepwise automated theorem proving system named MPS-Prover is proposed, which combines post-training data filtering and multi-perspective tree search techniques.
MR. Video: MapReduce as an Effective Principle for Long Video Understanding
Ziqi Pang (University of Illinois Urbana-Champaign), Yu-Xiong Wang (University of Illinois Urbana-Champaign)
RecognitionRetrievalOptimizationComputational EfficiencyTransformerLarge Language ModelPrompt EngineeringVision Language ModelVideoTextMultimodalityBenchmark
🎯 What it does: A long video understanding framework MR.Video based on the principles of MapReduce is proposed, which achieves detail perception and global reasoning of long videos through parallel processing of short segments and global aggregation;
MRO: Enhancing Reasoning in Diffusion Language Models via Multi-Reward Optimization
Chenglong Wang (Northeastern University), Tong Xiao (Northeastern University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelReinforcement LearningDiffusion modelText
🎯 What it does: This paper proposes a Multi-Reward Optimization (MRO) framework to enhance the performance of Diffusion Language Models (DLM) in reasoning tasks by strengthening the token correlations both within and between sequences during the decoding process.
MS-BART: Unified Modeling of Mass Spectra and Molecules for Structure Elucidation
Yang Han (Shanghai Jiao Tong University), Lu Chen (Shanghai Jiao Tong University)
TransformerLarge Language ModelSupervised Fine-TuningMultimodality
🎯 What it does: MS-BART is proposed, a unified language model framework for inferring molecular structures from mass spectrometry data.
MS-GS: Multi-Appearance Sparse-View 3D Gaussian Splatting in the Wild
Deming Li (Johns Hopkins University), Cheng Peng (Johns Hopkins University)
GenerationDepth EstimationGaussian SplattingPoint Cloud
🎯 What it does: Proposes the MS-GS framework to achieve high-quality 3D Gaussian point rendering under sparse views and multi-appearance scenarios.
msf-CNN: Patch-based Multi-Stage Fusion with Convolutional Neural Networks for TinyML
Zhaolan Huang (Freie Universität Berlin), Emmanuel Baccelli (Freie Universität Berlin)
OptimizationComputational EfficiencyConvolutional Neural NetworkImage
🎯 What it does: Developed the msf-CNN scheme, which utilizes multi-stage patch-based fusion to achieve dual optimization of memory and latency during CNN inference on MCUs.
MSTAR: Box-free Multi-query Scene Text Retrieval with Attention Recycling
Liang Yin (Huazhong University of Science and Technology), Yuliang Liu (Huazhong University of Science and Technology)
RetrievalTransformerVision Language ModelContrastive LearningImageTextBenchmark
🎯 What it does: MSTAR is proposed, a multi-query scene text retrieval method with box-free annotations, achieving unified retrieval for multiple query types.
MTL-KD: Multi-Task Learning Via Knowledge Distillation for Generalizable Neural Vehicle Routing Solver
Yuepeng Zheng (Shenzhen University), Yu Zhou (Eindhoven University of Technology)
OptimizationKnowledge DistillationTransformerReinforcement LearningTabular
🎯 What it does: Train a heavy decoder multi-task model through knowledge distillation to solve various VRP variants.
MTRec: Learning to Align with User Preferences via Mental Reward Models
Mengchen Zhao (South China University of Technology), Yi Cai (South China University of Technology)
Recommendation SystemReinforcement LearningSequential
🎯 What it does: By learning the user's psychological reward model, the sequential recommendation system is improved to better align with the user's true preferences.
Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search
Huanjin Yao (Tsinghua University), Dacheng Tao (Nanyang Technological University)
TransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningImageTextMultimodality
🎯 What it does: This study proposes the Collective Monte Carlo Tree Search (CoMCTS) method for learning-based reasoning in multimodal large language models (MLLMs) and constructs the Mulberry-260k dataset based on this method. Subsequently, the dataset is used for joint supervised fine-tuning of multimodal models, resulting in the Mulberry series models that possess o1-like step-by-step reasoning and reflection capabilities.
Multi-Agent Collaboration via Evolving Orchestration
Yufan Dang (Tsinghua University), Maosong Sun (Tsinghua University)
OptimizationLarge Language ModelReinforcement LearningText
🎯 What it does: A collaborative framework is proposed where a centralized 'puppeteer' dynamically directs multiple model agents to complete tasks.
Multi-Agent Debate for LLM Judges with Adaptive Stability Detection
Tianyu Hu, Tianlong Chen (University of North Carolina at Chapel Hill)
Large Language ModelAgentic AITextMultimodalityChain-of-Thought
🎯 What it does: A multi-agent debate framework is proposed, allowing multiple LLMs to collaboratively reason and improve judgment results through iterative communication; at the same time, an adaptive stability detection mechanism is designed to terminate the debate early when the judgment results tend to stabilize.
Multi-Agent Imitation by Learning and Sampling from Factorized Soft Q-Function
Yi-Chen Li (Nanjing University), Yang Yu (Nanjing University)
Robotic IntelligenceReinforcement LearningSequentialBenchmarkStochastic Differential Equation
🎯 What it does: The MAFIS framework is proposed for multi-agent imitation learning, combining soft Q function factorization with energy model sampling to achieve both distributed, online, and offline training modes.
Multi-agent KTO: Enhancing Strategic Interactions of Large Language Model in Language Game
Rong Ye (Fudan University), zhongyu wei
OptimizationTransformerLarge Language ModelReinforcement LearningTextChain-of-Thought
🎯 What it does: A multi-agent KTO (MaKTO) training framework is proposed, utilizing multi-model collaborative games to enable large language models to learn decision-making and dialogue simultaneously in social reasoning games like Werewolf, enhancing strategy and interaction effectiveness.
Multi-Agent Learning under Uncertainty: Recurrence vs. Concentration
Kyriakos Lotidis (Stanford University), Jose Blanchet (Stanford University)
Reinforcement LearningSequentialStochastic Differential Equation
🎯 What it does: This paper studies the convergence characteristics of multi-agent learning under uncertainty, analyzing the long-term behavior of two stochastic models' regularized learning in continuous games.
Multi-agent Markov Entanglement
Shuze Chen (Columbia University), Tianyi Peng (Columbia University)
Reinforcement LearningTabular
🎯 What it does: The concept of 'Markov Entanglement' is proposed, and it is proven to be a necessary and sufficient condition for value decomposition in multi-agent MDPs, providing the relationship between error upper bounds and entanglement measures;
Multi-Agent Reinforcement Learning with Communication-Constrained Priors
Guang Yang (Nanjing University), Yang Gao (Nanjing University)
Reinforcement Learning
🎯 What it does: A general communication constraint model is proposed to uniformly describe communication conditions in different scenarios, and based on this model, it distinguishes between lossy and lossless messages, thereby introducing a multi-agent reinforcement learning framework with communication constraints.
Multi-Class Support Vector Machine with Differential Privacy
Jinseong Park (Korea Institute for Advanced Study), Jaewook Lee (Seoul National University)
ClassificationSafty and PrivacyGaussian SplattingTabular
🎯 What it does: A differential privacy multi-class support vector machine (PMSVM) based on all-in-one is proposed, achieving privacy protection through a single data access.
Multi-dataset Joint Pre-training of Emotional EEG Enables Generalizable Affective Computing
Qingzhu Zhang (Southern University of Science and Technology), Quanying Liu (Southern University of Science and Technology)
ClassificationRecognitionContrastive LearningTime SeriesBiomedical Data
🎯 What it does: A multi-dataset joint pre-training framework mdJPT is constructed for emotional EEG recognition, achieving zero-shot/few-shot inference across datasets.
Multi-Environment POMDPs: Discrete Model Uncertainty Under Partial Observability
Eline M. Bovy (Radboud University Nijmegen), Nils Jansen (Ruhr University Bochum)
OptimizationReinforcement LearningTabularBenchmark
🎯 What it does: This paper studies the Multi-Environment POMDP (ME-POMDP) problem and proposes a framework for solving robust policies under uncertain discrete models.
Multi-Expert Distributionally Robust Optimization for Out-of-Distribution Generalization
Jinyong Jeong (Korea University), Seoung Bum Kim (Korea University)
Domain AdaptationOptimizationConvolutional Neural NetworkMixture of ExpertsImage
🎯 What it does: This paper proposes Multi-Expert Distributionally Robust Optimization (MEDRO), which equips each training environment with dedicated expert heads on a shared feature extractor and performs min-max DRO across all m² groups of expert-environment pairs, aiming to enhance robustness against discrete subpopulations and domain shifts.
Multi-head Temporal Latent Attention
Keqi Deng (University of Cambridge), Phil Woodland
CompressionComputational EfficiencyTransformerTextAudio
🎯 What it does: Designed and implemented Multi-Head Temporal Latent Attention (MTLA), a self-attention mechanism capable of dynamically compressing the time dimension of KV cache in the Transformer decoder, and evaluated its performance on various speech and text tasks.
Multi-head Transformers Provably Learn Symbolic Multi-step Reasoning via Gradient Descent
Tong Yang (Carnegie Mellon University), Yuejie Chi (Yale University)
TransformerGraphChain-of-Thought
🎯 What it does: This paper proves that a single-layer multi-head Transformer can learn symbolic multi-step reasoning tasks—finding paths in a tree (from target to root and from root to target)—using the Chain-of-Thought mechanism, and maintains zero error generalization capability on unseen tree structures through theoretical construction and gradient descent convergence analysis.
Multi-Kernel Correlation-Attention Vision Transformer for Enhanced Contextual Understanding and Multi-Scale Integration
Hongkang Zhang (Tsinghua University), Yanlong Wang (Tsinghua University)
ClassificationObject DetectionSegmentationTransformerSupervised Fine-TuningImage
🎯 What it does: This paper proposes the Multi-Kernel Correlation-Attention Vision Transformer (MK-CAViT), which extracts multi-scale features through three parallel convolution kernels (3×3, 7×7, 15×15) and implements cross-scale nonlinear dependency modeling using Fast-HGR correlation attention based on HGR.
Multi-modal contrastive learning adapts to intrinsic dimensions of shared latent variables
Yu Gui (University of Pennsylvania), Zongming Ma (Yale University)
Representation LearningConvolutional Neural NetworkContrastive LearningMultimodalityBiomedical Data
🎯 What it does: This study investigates the theoretical properties of multimodal contrastive learning (CLIP) under nonlinear representations and general data distributions, demonstrating that it can adaptively share the intrinsic dimensionality of latent variables through temperature optimization.
Multi-Modal Interactive Agent Layer for Few-Shot Universal Cross-Domain Retrieval and Beyond
Kaixiang Chen (Southeast University), Hui Xue (Southeast University)
ClassificationRetrievalAgentic AIVision Language ModelImageMultimodality
🎯 What it does: A general cross-domain retrieval (FS-UCDR) task for few-shot samples is proposed, along with a multi-modal interaction agent layer (MAIL) to enhance the cross-modal parameter update alignment of visual-language models, significantly improving retrieval and few-shot classification performance.
Multi-Modal View Enhanced Large Vision Models for Long-Term Time Series Forecasting
ChengAo Shen (University of Houston), Jingchao Ni (NEC Laboratories America)
TransformerVision Language ModelMultimodalityTime Series
🎯 What it does: A decomposition framework DMMV is proposed for long-term time series forecasting using multimodal perspectives and large visual models.
Multi-Objective Hyperparameter Selection via Hypothesis Testing on Reliability Graphs
Amirmohammad Farzaneh (King's College London), Osvaldo Simeone (King's College London)
OptimizationHyperparameter SearchText
🎯 What it does: A multi-objective hyperparameter selection framework RG-PT based on reliability graphs (DAG) is proposed, utilizing hypothesis testing to achieve FDR control;
Multi-Objective One-Shot Pruning for Large Language Models
Weiyu Chen (Yunnan Normal University), James Kwok
OptimizationLarge Language ModelText
🎯 What it does: This paper proposes a multi-objective one-shot pruning method called MOSP, which can generate sparse LLM models tailored to different user preferences, forming a Pareto front.
Multi-Objective Reinforcement Learning with Max-Min Criterion: A Game-Theoretic Approach
Woohyeon Byeon (Korea Advanced Institute of Science and Technology), Youngchul Sung (Korea Advanced Institute of Science and Technology)
OptimizationReinforcement Learning
🎯 What it does: A new framework for max-min multi-objective reinforcement learning (max-min MORL) is proposed, utilizing a game-theoretic perspective of two-player zero-sum games to transform the problem into learning Nash equilibria, followed by the design of a single-loop ERAM/ARAM algorithm to achieve synchronized updates of policies and weights.
Multi-order Orchestrated Curriculum Distillation for Model-Heterogeneous Federated Graph Learning
Guancheng Wan (Wuhan University), Mang Ye (Wuhan University)
Federated LearningKnowledge DistillationGraph Neural NetworkGraph
🎯 What it does: A knowledge distillation framework called TRUST is proposed for model heterogeneous federated graph learning, addressing the compatibility issue between different client models and the complexity of graph structures.
Multi-Scale Finetuning for Encoder-based Time Series Foundation Models
Zhongzheng Qiao (Nanyang Technological University), Savitha Ramasamy (Institute for Infocomm Research)
TransformerSupervised Fine-TuningTime Series
🎯 What it does: A multi-scale fine-tuning framework MSFT is proposed, which performs multi-scale fine-tuning on encoder-based time series models to enhance the predictive performance of downstream tasks.
Multi-scale Temporal Prediction via Incremental Generation and Multi-agent Collaboration
Zhitao Zeng (National University of Singapore), Yueming Jin (National University of Singapore)
SegmentationGenerationData SynthesisTransformerLarge Language ModelVision Language ModelDiffusion modelMultimodalityTime SeriesBenchmark
🎯 What it does: This paper proposes a multi-scale temporal prediction task, constructs a synchronized multi-scale annotation benchmark (MSTP), and introduces the IG-MC framework to achieve incremental generation of states and visuals, as well as multi-agent collaborative prediction.
Multi-step Visual Reasoning with Visual Tokens Scaling and Verification
Tianyi Bai (Hong Kong University of Science and Technology), Wentao Zhang (Peking University)
OptimizationReinforcement Learning from Human FeedbackTransformerLarge Language ModelSupervised Fine-TuningReinforcement LearningVision Language ModelImageMultimodality
🎯 What it does: A framework (VTS-V) is proposed that dynamically scales visual tokens during inference and performs validation, allowing multimodal large language models to iteratively select visual actions and evaluate their effectiveness through a validator.
Multi-Task Vehicle Routing Solver via Mixture of Specialized Experts under State-Decomposable MDP
Yuxin Pan (Hong Kong University of Science and Technology), Fangzhen Lin (Hong Kong University of Science and Technology)
OptimizationTransformerReinforcement LearningMixture of Experts
🎯 What it does: A unified multi-task vehicle routing problem (VRP) solver is proposed, utilizing the State-Decomposable MDP (SDMDP) and Latent-Space SDMDP (LS-SDMDP) frameworks to reuse benchmark VRP solvers and implement the Mixture-of-Specialized-Experts Solver (MoSES);
Multi-Token Prediction Needs Registers
Anastasios Gerontopoulos (Athena Research Center), Nikos Komodakis (IACM-Forth)
TransformerText
🎯 What it does: This paper proposes the MuToR method, which inserts learnable register tokens into autoregressive Transformer training to achieve multi-step prediction.
Multi-View Oriented GPLVM: Expressiveness and Efficiency
Zi yang, Pablo M. Olmos (Universidad Carlos III de Madrid)
Computational EfficiencyRepresentation LearningAuto EncoderImageText
🎯 What it does: A NG-MVLVM model for multi-view representation learning is proposed, which combines an expressive NG-SM kernel with random Fourier feature approximation;
Multiclass Loss Geometry Matters for Generalization of Gradient Descent in Separable Classification
Matan Schliserman (Tel Aviv University), Tomer Koren (Google Research)
ClassificationOptimization
🎯 What it does: In separable multi-class linear models, an upper bound on the generalization risk of gradient descent is provided under a finite number of iterations, and it is proven that the dependence of risk on the number of classes depends on the Lp smoothness of the loss template;
Multidimensional Bayesian Utility Maximization: Tight Approximations to Welfare
Kira Goldner (Boston University), Taylor Lundy (University of British Columbia)
Optimization
🎯 What it does: In the multidimensional Bayesian utility maximization problem with single-minded demand, the author proposes a class of mechanisms called Favorites (Random-Favorites and Prior-Free-Favorites) and proves that they can achieve utility maximization at a level close to social welfare in an i.i.d. single-minded demand environment; when the number of items is not less than the number of buyers, a constant approximation of (1-1/e) is obtained; when the number of buyers exceeds the number of items, a logarithmic approximation of O(log n/m) is achieved, and it is proven that this ratio is optimal in both cases; it also discusses why traditional revenue maximization tools struggle to obtain tighter upper bounds directly.
Multilevel neural simulation-based inference
Yuga Hikida (Aalto University), Francois-Xavier Briol
🎯 What it does: This paper proposes Multi-Layer Neural Likelihood Estimation/Posterior Estimation (ML-NLE/ML-NPE), which accelerates neural SBI by combining multi-layer Monte Carlo with low and high precision simulators.
Multimodal 3D Genome Pre-training
Minghao Yang (Hong Kong University of Science and Technology), Hui Xiong (Hong Kong University of Science and Technology)
TransformerContrastive LearningMultimodalityBiomedical Data
🎯 What it does: We propose the MIX-HIC multimodal foundational model, which jointly utilizes Hi-C contact maps and epigenomic trajectories for large-scale self-supervised pre-training, and applies it to various downstream tasks such as Hi-C prediction, chromosome loop detection, and CAGE expression.
Multimodal Bandits: Regret Lower Bounds and Optimal Algorithms
William Réveillard (KTH Royal Institute of Technology), Richard Combes (Université Paris-Saclay)
OptimizationReinforcement Learning from Human FeedbackGraph
🎯 What it does: This study investigates multi-armed bandits with a multi-peak structure and proposes a computable Graves-Lai optimization algorithm, achieving an asymptotically optimal bandit strategy.
Multimodal Causal Reasoning for UAV Object Detection
Nianxin Li (University of Electronic Science and Technology of China), Ce Zhu
Object DetectionVision Language ModelContrastive LearningImageMultimodality
🎯 What it does: In drone target detection, a framework called MCR-UOD based on multimodal causal reasoning is proposed, which enhances detection robustness through visual-language models and causal interventions.
Multimodal Disease Progression Modeling via Spatiotemporal Disentanglement and Multiscale Alignment
Chen Liu (Hong Kong Polytechnic University), Jing Qin (Hong Kong Baptist University)
ClassificationRecognitionTransformerImageMultimodalityBiomedical DataElectronic Health Records
🎯 What it does: The DiPro framework is proposed, which uses regional-level spatiotemporal separation and multi-scale alignment to integrate long sequences of chest X-rays and electronic medical records for disease progression modeling and ICU prediction.
Multimodal LiDAR-Camera Novel View Synthesis with Unified Pose-free Neural Fields
Weiyi Xue (Tongji University), Guang Chen (Tongji University)
Data SynthesisPose EstimationAutonomous DrivingNeural Radiance FieldMultimodalityPoint Cloud
🎯 What it does: A unified posture-free LiDAR-camera multimodal neural field framework MUP is proposed for posture-free novel view synthesis and pose estimation in large-scale scenes.
Multimodal Negative Learning
Baoquan Gong (Tianjin University), Bing Cao (Tianjin University)
ClassificationRecognitionMultimodality
🎯 What it does: A new Multimodal Negative Learning (MNL) framework is proposed, which allows weak modalities to suppress non-target categories under the guidance of strong modalities, thereby stabilizing the decision space and enhancing robustness.
Multimodal Tabular Reasoning with Privileged Structured Information
Jun-Peng Jiang (Nanjing University), Han-Jia Ye (Nanjing University)
OptimizationLarge Language ModelSupervised Fine-TuningReinforcement LearningMultimodalityTabular
🎯 What it does: Utilizing available structured tables as privileged information during training, the TURBO framework is proposed to enhance the multimodal reasoning capability for table images.
MultiNet: Adaptive Multi-Viewed Subgraph Convolutional Networks for Graph Classification
Xinya Qin (Beijing Normal University), Edwin Hancock
ClassificationGraph Neural NetworkGraphBiomedical Data
🎯 What it does: A MultiNet is designed and implemented, utilizing adaptive multi-view subgraph convolution and an alignment readout mechanism to address the issue of over-smoothing at the graph level and improve graph classification performance.
Multiplayer Federated Learning: Reaching Equilibrium with Less Communication
TaeHo Yoon (Johns Hopkins University), Nicolas Loizou (Johns Hopkins University)
Federated Learning
🎯 What it does: A new multi-player federated learning framework (MpFL) is proposed, modeling the federated learning process as a game among rational players, aiming to achieve equilibrium.
Multiplication-Free Parallelizable Spiking Neurons with Efficient Spatio-Temporal Dynamics
Peng Xue (Shenzhen Institute of Advanced Technology), Huihui Zhou (Shenzhen Institute of Advanced Technology)
Spiking Neural NetworkTime SeriesSequentialAudio
🎯 What it does: This paper proposes a multiplication-free, channel-wise separable, and parallelizable spiking neuron model called mul-free channel-wise PSN, aimed at enhancing the learning performance of temporal data.
Multipole Attention for Efficient Long Context Reasoning
Coleman Richard Charles Hooper (University of California), Amir Gholami (University of California)
Computational EfficiencyTransformerLarge Language ModelText
🎯 What it does: The Multipole Attention method is proposed, which uses clustered key centroids to apply precise attention to important keys, while the remaining keys are represented approximately to accelerate long-chain inference.
Multiresolution Analysis and Statistical Thresholding on Dynamic Networks
Raphael Romero, Alexander Modell (Imperial College London)
Anomaly DetectionGraph Neural NetworkGraphTime Series
🎯 What it does: This paper proposes the ANIE method for adaptively identifying structural changes at different time scales in continuous-time dynamic networks.
MultiScale Contextual Bandits for Long Term Objectives
Richa Rastogi (Cornell University), Thorsten Joachims (Cornell University)
Recommendation SystemReinforcement LearningVideoText
🎯 What it does: A multi-scale strategy learning framework is proposed, which constructs hierarchical priors using rich short-term feedback at lower levels under multi-level time scales, and trains context-independent multi-level Bandit strategies;
Multiscale guidance of protein structure prediction with heterogeneous cryo-EM data
Rishwanth Raghu (Princeton University), Ellen D Zhong
Protein Structure PredictionDiffusion modelPoint Cloud
🎯 What it does: A multi-scale guidance method called CryoBoltz is proposed, which utilizes heterogeneous cryo-EM data to guide a pre-trained diffusion structure prediction model, enabling the model to generate atomic models that conform to different conformations under experimental data constraints.
Multitask Learning with Stochastic Interpolants
Hugo Negrel (Capital Fund Management), Eric Vanden-Eijnden (Capital Fund Management)
RestorationGenerationFlow-based ModelImageStochastic Differential EquationOrdinary Differential Equation
🎯 What it does: A multi-task generative framework based on operator interpolation is proposed, utilizing linear operators to replace traditional time variables, constructing a universal generative model that can seamlessly switch between different tasks.
Multivariate Latent Recalibration for Conditional Normalizing Flows
Victor Dheur (University of Mons), Souhaib Ben Taieb (Mohamed bin Zayed University of Artificial Intelligence)
GenerationFlow-based ModelImageTabular
🎯 What it does: A Latent Recalibration (LR) method is proposed for post-calibration in the latent space of conditional reversible generative models (such as normalizing flow and flow matching);
Multivariate Time Series Anomaly Detection with Idempotent Reconstruction
Xin Sun (Zhejiang University), Chao Li (Zhejiang University)
Anomaly DetectionAuto EncoderGenerative Adversarial NetworkTime Series
🎯 What it does: This paper proposes and validates a pluggable Idempotent Generation Module (IGAD) to improve multivariate time series anomaly detection models.
Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generation
Xinyu Yang (Carnegie Mellon University), Beidi Chen (Carnegie Mellon University)
GenerationComputational EfficiencyTransformerLarge Language ModelText
🎯 What it does: The Multiverse framework is proposed, achieving adaptive Map-Reduce parallel generation and efficient inference on a 32B model.
MuRating: A High Quality Data Selecting Approach to Multilingual Large Language Model Pretraining
Zhixun Chen (Hong Kong University of Science and Technology), Meng Fang (University of Liverpool)
TransformerLarge Language ModelContrastive LearningText
🎯 What it does: Proposes the MuRating framework, which aggregates various English automatic evaluators and transfers them to multilingual data quality assessment.
MURKA: Multi-Reward Reinforcement Learning with Knowledge Alignment for Optimization Tasks
Wantong Xie (University of Science and Technology of China), Xiangyang Li
OptimizationKnowledge DistillationReinforcement LearningText
🎯 What it does: MURKA is an end-to-end multi-agent framework that automatically converts natural language descriptions into executable optimization models.
MuSLR: Multimodal Symbolic Logical Reasoning
Jundong Xu (National University of Singapore), Wynne Hsu (National University of Singapore)
TransformerLarge Language ModelVision Language ModelTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: This paper proposes the multi-modal symbolic logic reasoning (MuSLR) task and constructs the corresponding benchmark dataset MuSLR-Bench; it also introduces a LogiCAM framework based on chunked reasoning.
MUSTAFAR: Promoting Unstructured Sparsity for KV Cache Pruning in LLM Inference
Donghyeon Joo (University of Maryland), Bahar Asgari (d-Matrix)
TransformerLarge Language ModelTextBenchmark
🎯 What it does: Pruning the Key/Value cache of LLM using unstructured sparsification, combined with bitmap compression format and custom sparse attention kernels, significantly reduces KV cache usage and improves inference throughput.
MutualVPR: A Mutual Learning Framework for Resolving Supervision Inconsistencies via Adaptive Clustering
Qiwen Gu (Tongji University), Guang Chen (Tongji University)
RetrievalRepresentation LearningTransformerContrastive LearningImage
🎯 What it does: This paper proposes MutualVPR, a visual localization framework that achieves mutual learning through adaptive clustering to address the issue of supervision inconsistency caused by viewpoint changes and occlusions.
MV-CoLight: Efficient Object Compositing with Consistent Lighting and Shadow Generation
Kerui Ren (Shanghai Jiao Tong University), Bo Dai (The University of Hong Kong)
GenerationData SynthesisComputational EfficiencyTransformerGaussian SplattingImage
🎯 What it does: This paper proposes a two-stage feed-forward framework MV-CoLight for achieving consistent lighting and realistic shadows in object synthesis within 2D images and 3D scenes, supporting multi-view real-time rendering.
MVSMamba: Multi-View Stereo with State Space Model
Jianfei Jiang (University of Science and Technology Beijing), Huimin Ma (University of Science and Technology Beijing)
Depth EstimationConvolutional Neural NetworkImage
🎯 What it does: MVSMamba is proposed, a multi-view stereo reconstruction network based on the Mamba architecture;
Mysteries of the Deep: Role of Intermediate Representations in Out of Distribution Detection
Imezadelajara, Ehsan Abbasnejad (Australian Institute for Machine Learning)
Anomaly DetectionTransformerVision Language ModelImage
🎯 What it does: This paper studies the enhancement of detection performance in zero-shot OOD detection by utilizing intermediate layer representations from pre-trained vision-language models (such as CLIP);
Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards
Qingming LIU, Kui Jia (Chinese University of Hong Kong)
GenerationReinforcement LearningDiffusion modelPoint CloudStochastic Differential Equation
🎯 What it does: By using a 2D differentiable reward model for reinforcement learning fine-tuning on a 3D local diffusion model, we propose Nabla-R2D3 to achieve efficient alignment.
NaDRO: Leveraging Dual-Reward Strategies for LLMs Training on Noisy Data
Haolong Qian (Tsinghua University), Chun Yuan (Microsoft)
TransformerLarge Language ModelReinforcement LearningText
🎯 What it does: The research trains large language models for long-sequence decision-making under noisy data and proposes the NaDRO framework.
Native Segmentation Vision Transformers
Guillem Braso, Laura Leal-Taixé (NVIDIA)
SegmentationTransformerImage
🎯 What it does: A visual Transformer backbone network capable of self-generating segmentation masks is designed, utilizing spatial grouping layers to achieve end-to-end learnable hierarchical segmentation;
Native-Resolution Image Synthesis
ZiDong Wang, Yiyuan Zhang (Chinese University of Hong Kong)
GenerationData SynthesisTransformerDiffusion modelImage
🎯 What it does: The concept of Native-resolution image synthesis is proposed, training a single model to generate high-quality images at arbitrary resolutions and aspect ratios.
Natural Gradient VI: Guarantees for Non-Conjugate Models
Fangyuan Sun (ETH Zurich), Niao He (ETH Zurich)
OptimizationTabular
🎯 What it does: This paper studies the theoretical properties of Natural Gradient Variational Inference (NGVI) under non-conjugate models and proposes the Proj-SNGD algorithm based on relative smoothness, achieving global convergence guarantees.
NAUTILUS: A Large Multimodal Model for Underwater Scene Understanding
Wei Xu (Huazhong University of Science and Technology), Xiang Bai (Huazhong University of Science and Technology)
ClassificationObject DetectionDepth EstimationTransformerLarge Language ModelSupervised Fine-TuningVision Language ModelImageTextMultimodality
🎯 What it does: The NAUTILUS large multimodal model (LMM) is proposed for underwater scene understanding, and the first underwater instruction-following dataset NautData is constructed, covering three levels of granularity and eight tasks, including coarse and fine classification, counting, VQA, detection, localization, area description, and image description.
NavBench: Probing Multimodal Large Language Models for Embodied Navigation
Yanyuan Qiao (Swiss Federal Institute of Technology Lausanne), Qi Wu (Swiss Federal Institute of Technology Lausanne)
Robotic IntelligenceTransformerLarge Language ModelVision-Language-Action ModelTextMultimodalityBenchmarkChain-of-Thought
🎯 What it does: We propose NavBench, a zero-shot evaluation benchmark for multimodal large language models in embodied navigation, covering two main modules: cognitive understanding (global instruction matching, progress estimation, local observation-action reasoning) and step execution (graded by difficulty), and providing a practical real-world deployment pipeline.
Navigating the MIL Trade-Off: Flexible Pooling for Whole Slide Image Classification
Hossein Jafarinia (Sharif University of Technology), Mohammad Hossein Rohban (Sharif University of Technology)
ClassificationTransformerImageBenchmark
🎯 What it does: This paper addresses WSI classification under low data conditions, re-examining and improving traditional MIL aggregation methods, proposing Maxsoft pooling and PerPatch augmentation;
NaViL: Rethinking Scaling Properties of Native Multimodal Large Language Models under Data Constraints
Changyao Tian (Shanghai AI Laboratory), Jifeng Dai (Tsinghua University)
TransformerLarge Language ModelMixture of ExpertsVision Language ModelImageTextMultimodality
🎯 What it does: The research focuses on end-to-end training of multimodal large language models under data-constrained conditions, systematically exploring the design space and scaling characteristics, and proposing the NaViL model.
Near-Exponential Savings for Population Mean Estimation with Active Learning
Julian Morimoto, Daniel E. Ho (Stanford University)
OptimizationComputational EfficiencyBiomedical DataElectronic Health Records
🎯 What it does: A two-stage active learning algorithm named PartiBandits is proposed to estimate the mean of multi-class random variables under a limited labeling budget.
Near-Optimal Experiment Design in Linear non-Gaussian Cyclic Models
Ehsan Sharifian (École Polytechnique Fédérale de Lausanne), Negar Kiyavash (École Polytechnique Fédérale de Lausanne)
OptimizationGraph
🎯 What it does: This study investigates the problem of causal structure learning in linear non-Gaussian structural equation models, proposing an approximately optimal experimental design method that combines observational and interventional data.
Near-Optimal Quantum Algorithms for Computing (Coarse) Correlated Equilibria of General-Sum Games
Tongyang Li (Peking University), Yexin Zhang (Peking University)
OptimizationReinforcement Learning from Human Feedback
🎯 What it does: An algorithm is proposed for solving the ε-correlated equilibrium (CE) and ε-coarse correlated equilibrium (CCE) of multi-player general summation games based on quantum algorithms.
Near-Optimal Regret-Queue Length Tradeoff in Online Learning for Two-Sided Markets
Zixian Yang (University of Michigan), Lei Ying (University of Michigan)
Reinforcement Learning
🎯 What it does: This study investigates a matching queue model for price-sensitive customers and service providers in two-sided markets, designing online learning pricing and matching strategies.
Near-Optimal Sample Complexity for Online Constrained MDPs
Chang Liu (University of California Los Angeles), Lin Yang
OptimizationReinforcement Learning
🎯 What it does: This paper proposes a model-based dual iterative algorithm for online learning of constrained Markov decision processes (CMDP) in finite state spaces, and provides the approximate optimal sample complexity under relaxed and strictly feasible conditions.
Nearly Dimension-Independent Convergence of Mean-Field Black-Box Variational Inference
Kyurae Kim (University of Pennsylvania), Jacob R. Gardner (University of Pennsylvania)
Optimization
🎯 What it does: The paper conducts a theoretical analysis of the convergence speed of the mean-field location-scale family in black-box variational inference (BBVI), proving that under strongly log-concave and smooth objectives, the iterative complexity does not significantly increase with the growth of dimensions.
Nearly-Linear Time and Massively Parallel Algorithms for $k$-anonymity
Kevin Aydin (Google Research), Peilin Zhong (Google Research)
Safty and PrivacyComputational EfficiencyTabular
🎯 What it does: A k-anonymization suppression algorithm based on nearest neighbor clustering is proposed, providing an O(k) approximate solution in near-linear time, along with its implementation under the large-scale parallel computing model (MPC).
Nearly-Linear Time Private Hypothesis Selection with the Optimal Approximation Factor
Maryam Aliakbarpour (Rice University), Vincent X. Wang (Rice University)
Safty and PrivacyComputational Efficiency
🎯 What it does: This paper proposes a near-linear time hypothesis selection algorithm under the central differential privacy model, capable of completing the hypothesis selection task with a logarithmic sample size, approximately linear time, and an approximation factor of 3.